A Compelling Example of AI in CAD: Autodesk’s Take on Generative Design18 Feb, 2019 By: Alex Herrera
Herrera on Hardware: Both evolutionary and revolutionary uses are emerging for artificial intelligence (AI) in CAD, which is making a splash from NVIDIA rendering applications to Autodesk Fusion 360 — and suggesting serious synergy with virtual workstations.
Arriving at the right model, or at least the best one to explore first, shouldn’t end at a creating a conceptual, stand-alone structural representation. Rather, the key is to also allow the means to bridge the gap from that machine-generated model to engineering verification and styling, and eventually on to getting it manufactured. A simple mesh STL-type output is of limited value; it yields an interesting shape, but one that can’t serve any use beyond the visual. A productive workflow needs that geometry embedded with all the rich metadata needed to directly feed into simulation, verification, rendering, and on to prototype and manufacturing. If the machine synthesis produced just a shell of geometry, the designer might have to re-create the whole thing manually. Bridging the gap from synthesis to the rest of the workflow was a key design goal for Fusion 360’s generative design functionality, and Autodesk says its implementation is unique in the way it automates the synthesized design in a complete, usable, and editable format ready for verification, re-design, and ultimately, physical creation.
A generative design–based workflow should seamlessly culminate in a manufacturable model, be it for prototype or volume production. As such, Fusion 360’s generative design considers the constraints and capabilities of the manufacturing process and materials available. That knowledge and attention is particularly relevant in the context of additive methods like 3D printing, especially now given recent advancements in printing with metal. Such methods allow for shapes and structure that conventional manufacturing methods can’t achieve, making it essential for the user to provide Fusion 360’s generative design the guidance on which methods are available or preferred.
Synthetic models generated to meet the same constraints, but targeted for different manufacturing methods. Image source: Autodesk
The What and Where Don’t Concern the End User
For now, Autodesk has AI-based generative design focused exclusively in Fusion 360, a purely cloud-based solution. And that makes a lot of sense when considering the emerging landscape of both cloud and workstation computing. With a cloud implementation like Fusion 360, all computes, graphics, and data are colocated in a common datacenter. (Sound like a familiar paradigm? It’s essentially the cloud version of the virtual workstation you’ve heard about, including in this column. For more information, see my series, “Harnessing the Cloud for CAD: The Case for Virtual Workstations.”)
The two approaches share obvious synergy, and the appeal of cloud-based generative design should only further spur the motivation of professionals seeking to transition their workstations to the cloud. Cloud providers including Amazon Web Services (AWS), Microsoft Azure, and Google Cloud aggressively create and deploy machine instances that deliver physical or virtual/shared resources that make even the most capable deskside workstation seem wimpy by comparison. Consider the performance of a machine equipped with 96 CPUs and eight top-end NVIDIA AI-accelerating Volta GPUs could do when paired with 768 GB of memory and dedicated 1.8 TB of NVMe SSD. That set of hardware is certainly not going to reside at your desk, but it potentially could be harnessed for generative design by an application like Fusion 360, or by the rent-as-you-go virtual workstation your company eventually deploys instead of traditional machines.
Examples of available Amazon Web Services EC2 machine instances. Data source: AWS.
The instances above represent just a few of the many machine instances that providers such as AWS make available for use in the cloud (and some which Fusion 360 may use, but not necessarily). Some are intentionally optimized for compute-bottlenecked loads, while others are optimized for memory or storage. Some include GPU resources for help with appropriate workloads, which can include AI and workloads such as generative design. (Not shown, but commensurate in magnitude, are available storage and network bandwidth.)
Of course, neither Autodesk nor suppliers like Intel, AMD, and NVIDIA are going to ask or expect end users to navigate the appropriate performance, features, virtual machine types, or determine when CPUs or GPUs might best for generative design (or any other) workloads headed to the cloud for execution. As you might guess, Autodesk hides the processing wizard behind the curtain so the user needn’t be concerned with how it gets done. Based on the model and constraints, Fusion 360 assesses the workload at hand, “right-sizes” it to the appropriate machine instances, and when complete, hands it back to you inside the application. The user doesn’t care what is performing the synthesis, nor where — which is precisely one of the major appeals of cloud computing.
Compelling Uses Are Already Here, with Many More to Come
Over time, machine learning will permeate virtually every corner of computing technology and applications. Of that, the majority of us have little doubt. In CAD computing, uses have already popped up to significantly improve the performance of existing 3D graphics and rendering that professionals require. But most certainly, those more evolutionary uses represent but the tip of the iceberg, and the real impact will come in more revolutionary applications.
The demands and workloads CAD represent make it fertile ground for AI infiltration. Expect machine learning advancements to both leverage and transform the existing tried-and-true design/verify/iterate/manufacture workflow. Most of the possibilities we’ve likely not yet imagined, but the same would likely have been said ten years ago for generative design, an approach that offers compelling, undeniable appeal. Combine that with its natural synergy with virtual workstations, and we should see momentum increase for both approaches.
It’s time to pay close attention to what AI and the cloud offer for CAD computing — assisting in design creation itself, leveraging the technology to streamline workflows, improving end products, cutting costs, and shortening time to markets. Your competition likely has.